Session: RIP: Predictive Modeling and Characterization of Corrosion Processes in Complex Environments (In Honor of Professor Digby Macdonald) (Part I of IV)
Physics-Informed Neural Network Point Defect Model for Predicting Passive Film Breakdown in High-Entropy Alloys (RIP2026-00050)
The stability of passive oxide films is a critical factor in determining the corrosion resistance of structural alloys, particularly in chloride-containing environments. This study presents the integration of the Refined Point Defect Model (R-PDM) with physics-informed neural networks (PINNs) to predict the electrochemical behavior of passive films and their breakdown dynamics in high entropy alloys (HEAs). The framework integrates defect transport theory with machine learning, coupling vacancy generation, migration, and annihilation with electrochemical kinetics at the metal/film and film/solution interfaces. The model incorporates both chloride ion-induced film dissolution and penetration mechanisms, thereby addressing two long-standing hypotheses in corrosion science. A multi-objective PINN training strategy enforces fidelity to experimental data while simultaneously satisfying governing physical constraints, including defective transport equations, boundary flux conditions, and current balance laws. Validation against electrochemical measurements and voltammetry simulations demonstrates the accurate prediction of open-circuit potential (OCP), critical breakdown potential (Eb), and steady-state film thickness across diverse alloy systems. Application to high-entropy alloys (HEAs) highlights the model’s ability to map compositional complexity to kinetic parameters, enabling predictive insights into alloy-specific passivity and breakdown behavior. Results reveal that increasing chloride concentration accelerates film thinning, enhances electric field strength, and shifts Ecorr toward more negative values, consistent with experimental observations. This research advances the predictive modeling of corrosion by uniting rigorous electrochemical theory with modern machine learning. The proposed framework provides a pathway toward the rational design of next-generation alloys with tailored passivity and enhanced durability in aggressive environments.
